5
\$\begingroup\$

I am trying to create a very simple timer framework which allows you to setup event handling based on a timeout.

The basic primitives allow the programmer to:

  • allocate or free timers
  • arm/disarm timers
  • set attributes and timeout for a timer

Currently, the self contained example will create timers up to the maximum (currently defined as 10) with random timeout values from up to 20 seconds. Each timer will eventually expire and the associated call-back function executed.

          /* Very simplistic timer framework by amallory@qnx.com */
#include <stdio.h>
#include <unistd.h>
#include <stdlib.h>
#include <time.h> #include <sys/time.h>
#include <inttypes.h>
#include <sys/queue.h>

#define NUM_TIMERS 10
#define MAX_RANDOM_TIME_MS  20000

enum timer_callback_retval {
    CB_RETURN_NORMAL = 0,
    CB_RETURN_FREE_TIMER,
    CB_RETURN_INVALID,
};

enum timer_type {
    TT_RELATIVE = 0,
    TT_ABSOLUTE,
    TT_INVALID,
};

/* Define our timer list type */
TAILQ_HEAD(timer_list, timer_node);

/* Master tick clock/count */
uint64_t tick_cnt = 0;

/*
Timer data structure:
-the linked list
-the monotonic fire time (saved as an absolute time)
-the user callback handler to run on expiry of timer
-the registered data pointer to pass to the user callback
*/
struct timer_node {
    TAILQ_ENTRY(timer_node) entries;
    uint64_t fire;
    int (*cb)(void* user_data);
    void *user_data;
};

/* Our global timer lists */
struct timer_list active_timers;
struct timer_list free_timers;
struct timer_node *timer_memory;

#ifdef DEBUG
/* print out the contents of a given timer list */
void print_list(struct timer_list *list) {
    struct timer_node *np;
    TAILQ_FOREACH(np, list, entries) {
        printf("timer fire %llu\n", np->fire);
    }
}
#endif

/* put a timer onto the free list */
static void free_timer(struct timer_node *timer) {
    TAILQ_INSERT_HEAD(&free_timers, timer, entries);
}

/* pull an available timer off the free list */
static struct timer_node* alloc_timer(void) {
    struct timer_node *np;
    if(TAILQ_EMPTY(&free_timers)) return NULL;

    np=TAILQ_FIRST(&free_timers);
    TAILQ_REMOVE(&free_timers, np, entries);
    return np;
}

/* put a timer onto the actives timer queue */
static void arm_timer(struct timer_node* timer) {
    struct timer_node *np;

    if(TAILQ_EMPTY(&active_timers)) {
        TAILQ_INSERT_HEAD(&active_timers, timer, entries);
    } else {
        for(np=TAILQ_FIRST(&active_timers) ; np ; np=TAILQ_NEXT(np, entries)) {
            if(timer->fire < np->fire) {
                TAILQ_INSERT_BEFORE(np, timer, entries);
                return;
            }
        }
        TAILQ_INSERT_TAIL(&active_timers, timer, entries);
    }
}

/* remove a timer from the actives timer queue */
static void disarm_timer(struct timer_node* timer) {
    TAILQ_REMOVE(&active_timers, timer, entries);
return;
}

/*  set timer attributes such as relative/absolute fire timer,
fire timer, callback and user data passed to callback
*/
static int set_timer(struct timer_node* timer, enum timer_type tt, uint64_t fire,    int (*cb)(void*), void* user_data) {
    switch(tt) {
    case TT_RELATIVE:
            fire+=tick_cnt;
            break;
        case TT_ABSOLUTE:
            break; /* do nothing */
        case TT_INVALID:
        default:
            return -1;
    }
    timer->fire = fire;
    timer->cb = cb;
    timer->user_data = user_data;
    return 0;
}


/* initialisation of the timer subsystem */
static void init_timers(void) {
    unsigned i;

    TAILQ_INIT(&active_timers);
    TAILQ_INIT(&free_timers);

    /*  We're preallocating the memory and using a fixed timer pool size to keep
    things simpler and avoid cluttering this exercise with lots of error checking
    for bad memory conditions.  This works or we're toast.
    */
    if((timer_memory = malloc(sizeof(struct timer_node)*NUM_TIMERS)) == NULL) {
        perror("Fatal! Can't allocate our block of timers!");
        exit(EXIT_FAILURE);
    }

    /* seed our free timer list */  
    for(i=0 ; i < NUM_TIMERS; i++) {
        free_timer(&timer_memory[i]);
    }
}

/* Our clock handling routine that runs each clock tick */
static void clock_tick(int signo) {
    struct timer_node *np;

    tick_cnt++; 
    while(!TAILQ_EMPTY(&active_timers) && (np=TAILQ_FIRST(&active_timers)) && np->fire  <= tick_cnt) {
        disarm_timer(np);
        if(np->cb(np->user_data) == CB_RETURN_FREE_TIMER) free_timer(np);
    }

}

/*  Setup a simulated clock tick that functions much like a real clock
interrupt might.  We're using *NIX signals and a process timer as
it is a very common service available on almost any *NIX like system.
While not perfect, it's good enough to be illustrative.
*/
static int init_ticker(unsigned ms) {
    struct itimerval it;
    struct timeval tv;

    tv.tv_sec = 0;
    tv.tv_usec = ms*1000;

    it.it_interval = it.it_value = tv;
    signal(SIGALRM, clock_tick);
    return setitimer(ITIMER_REAL, &it, NULL);
}

/* The user timer callback function */
int tcb(void *data) {
    struct timer_node *np = data;

    /*  Normally you wouldn't only printf() as a result of a timer
    but it is sufficient to be illustrative.
*/  
    printf("Timer Callback : %llu\n", np->fire);
    return CB_RETURN_FREE_TIMER;
}

int main(int argc, char* argv[]) {
    struct timer_node *np;
    int i;

    init_timers(); /* init the timer subsystem */
    init_ticker(1); /* 1ms tick simulating a hw clock */

    /* Create a bunch of timers from 1 to 5000ms in time and arm them */
    for (i=0 ; i < NUM_TIMERS ; i++) {
        if((np = alloc_timer()) == NULL) {
            perror("Fatal! we ran out of timers?");
            exit(EXIT_FAILURE);
        }

        if(set_timer(np, 0, (rand()+1) % MAX_RANDOM_TIME_MS , tcb, np) == -1) {
            perror("Fatal! Bad timer set!");
            exit(EXIT_FAILURE);
    }
        arm_timer(np);
    }

    /* Sit around letting the timers expire - not pretty but simple */
    while(1) {
        sleep(100);
    }

    /* never reached */

    return 0;
}

I would like to identify as many weaknesses or bugs in the framework as I can find. You are welcome to identify any type of issue you feel is relevant. Anything from logic errors to algorithmic limitations as this is important for me to know all its weaknesses and make sure this piece of code runs efficient, bug free and that I am using best coding practices.

\$\endgroup\$

migrated from stackoverflow.com Jan 3 '12 at 5:43

This question came from our site for professional and enthusiast programmers.

  • 2
    \$\begingroup\$ Your signal handler should be async-safe see securecoding.cert.org/confluence/display/seccode/… so your code is utterly wrong in its design \$\endgroup\$ – Basile Starynkevitch Jan 3 '12 at 5:34
  • \$\begingroup\$ One specific issue is that you are running your user functions in a signal handler. This is very bad as there are very few safe functions that can be called in a signal handler. Move the receiving of the signals and calling of the functions onto another thread. You are also not reestablishing your signal handler, use sigaction instead. \$\endgroup\$ – Lalaland Jan 3 '12 at 5:36
  • \$\begingroup\$ Thats Great guys, please keep your comments coming! \$\endgroup\$ – user1127217 Jan 3 '12 at 19:12
3
\$\begingroup\$

Usability could be improved a bit. For example, alloc_timer() (note, name is deceptive as no actual allocation takes place) could be put inside of the set_timer() function. Let your framework worry about finding a free timer in the pool, or returning an error code. A separate function for allocating the timer would be more useful if you weren't using a pool and if it was important for a user to control when memory allocations occur.

I don't like how the callback method controls whether the timer is freed. I could see lots of potential timer 'leaks'. I think if you added a repeating timer type, most of the cases where you wouldn't want to free your timer could then be taken care of by your framework. If you really need a handle to a timer that you could guarantee would be available when you need it, I think extending or overloading alloc_timer() (and/or set_timer()) would be a better place. That way, the decision to take it off and put it back on the free_timers list is in the same line of code.

In clock_tick(), you're removing an element off the active_timers list while you iterate over it. In your case it looks like it'll work but that's usually an unsafe thing to do, I'd double check that.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.